Data Scientist

- Built a CUDA-accelerated pipeline using Whisper and quantized foundational LLMs for call transcription and summarization; integrated into the company’s CRM to extract structured insights. - Generated debtor embeddings from transcripts, CRM notes, and legal documents using HuggingFace models. Indexed with FAISS and a RAG pipeline facilitated contextual retrieval. - Developed a binary restructuring model in a highly imbalanced setting, combining tabular and text features. Tracked with MLflow, delivered via CI/CD/CT, and monitored with Prometheus. Used SHAP and DoWhy for explainability and causal analysis. - Designed sentiment models to analyze voice tone in debtor call communications to improve agent outreach and escalation strategies. - Built AVM models enriched with geolocation features to improve property valuation accuracy and support portfolio reporting and asset prioritization.

- Built and deployed a VIP conversion model from early user behavior in an extremely imbalanced setting; enabled revenue-based segmentation and targeted offers. - Developed a distributed anomaly detection pipeline with inter-metric root-cause analysis using Prophet to monitor daily product and operational KPIs and accelerate diagnostics. - Designed sentiment models over customer support chats to assess tone and identify churn risks, supporting customer experience improvements and targeted follow-ups. - Built a two-step customer segmentation pipeline for the Sportsbook product using BIRCH and hierarchical clustering, providing business teams with interpretable personas for product development and campaign design.

- Developed ensemble models to optimize property acquisition, prioritizing high-ROI opportunities and improving capital allocation across markets. - Deployed dynamic pricing and elasticity models to calibrate booking rates, reducing marketing costs by 7%. - Built an automated invoice processing system using OCR and Airflow-managed ETL, reducing manual workload and increasing reliability in financial workflows. - Created NPS prediction models with SHAP to reveal satisfaction drivers and guide product improvements.

- Delivered customer segmentation pipelines (CNN + XGBoost) that increased retention by 15% through behavior-based targeting. - Built a real-time analytics stack (Kafka, Druid, Divolte) and designed GCP-based data lake architecture to support scalable analytics. - Developed pricing and competitor tracking tools via automated web scraping, surfacing market insights for strategic decisions.